Add files via upload

pred.py

@@ -0,0 +1,137 @@
+from paddleocr import PaddleOCR
+import cv2
+from PIL import Image, ImageDraw, ImageFont
+import numpy as np
+import os
+import matplotlib.pyplot as plt
+
+ocr = PaddleOCR(
+    use_angle_cls=True,
+    lang='en',
+    drop_score=0.35,
+    det_db_score_mode='fast',
+    max_text_length=6,
+    rec_batch_num=8,
+    rec_algorithm="CRNN", 
+    det_db_box_thresh=0.5,
+    det_db_thresh=0.35,
+    det_db_unclip_ratio=1.9
+
+    # rec_char_dict_path= "./ppocr/utils/ppocr_keys_v1.txt"
+) # need to run only once to download and load model into memory
+
+# input and output paths
+input_path = './content/inputs'
+output_path = './content/outputs'
+
+# Create the directories if they do not exist
+if not os.path.exists(input_path):
+    os.makedirs(input_path)
+
+if not os.path.exists(output_path):
+    os.makedirs(output_path)
+
+labels = []
+preds = []
+
+dict_replace = {"o":"0", "O":"0", "D":"0", "a":"0", "e":"0", "I":"1", "l":"1", "i":"1", "z":"2", "Z":"2", "y":"4", "L+":"4", "L":"4", "h":"4", "H":"4", "u":"4", "V":"4",
+                 "S":"5", "S":"5", "b":"6", "G":"6", "F":"7", "J":"7", "t":"7", "B":"8", "g":"9", ":":".", ",":".", "-":".", "..":"."}
+
+
+def post_processing(text):
+    # POSTPROCESSING
+    # Replacing strings and sembols to similar numbers
+    for k, v in dict_replace.items():
+        text = text.replace(k, v)
+    if len(text)>=3:
+        if text[-3] not in ["."]:
+            text = text[:-2] + "." + text[-2:] # insert "." to the position of -3
+
+    if text[-1] != '0':
+        text = text[:-1] + '0' # replace the last character with '0'
+
+    if len(text) < 2:
+        text = text + '.00' # if lenght is smaller than 2 add '.00' to the output
+
+    if len(text) < 3:
+        text = text + '.00' # if lenght is smaller than 3 add '.00' to the output
+
+    if text[-2] == '.':
+        text = text[:-2] + '0' + text[-1] # if text[-2] == '.' change its value to '0' (most possible value)
+
+    if len(text) > 2:
+        dot_index = text.find('.')
+        if dot_index != len(text)-3 and dot_index != -1:
+            text = text[:dot_index] + '0' + text[dot_index+1:]
+    print("text: ", text)
+
+    check_first = text.split('.')[0]
+    check_last = text.split('.')[1]
+    print("first: ", check_first)
+    print("last: ", check_last)
+    if len(check_first) > 3 and len(check_last) == 2:
+        lenght = len(check_first)
+        index = lenght - 3
+        check_first = check_first[:-index]
+        text = check_first + '.' + check_last
+
+    return text
+
+
+for filename in os.listdir(input_path):
+    actual_label = filename.split("_")[0]
+    actual_label = actual_label.replace(":", ".").replace(",", ".").replace("-", ".")
+    print(f"Actual label is : {actual_label}")
+    labels.append(actual_label)
+
+    image_path = os.path.join(input_path, filename)
+    img = cv2.imread(image_path, cv2.IMREAD_UNCHANGED)
+    img = np.array(img) # convert to numpy array
+    result = ocr.ocr(img, cls=False) # cls makes rotations for prediction if set True
+
+    # Draw bounding boxes and OCR result on the image
+    image = Image.fromarray(img)
+    draw = ImageDraw.Draw(image)
+    font = ImageFont.truetype("./StyleText/fonts/en_standard.ttf", 30) # modify the font size here
+
+    for res in result:
+        if len(res) > 0:
+            text = res[0][1][0] # get the OCR result text
+            text = post_processing(text)
+
+            print(f"prediction : {text}")
+            text += ' (' + str(round(res[0][1][1]*100, 2)) + '%)'
+            bbox = res[0][0]
+
+            # Convert list of four points to tuple of two points
+            bbox = ((bbox[0][0], bbox[0][1]), (bbox[2][0], bbox[2][1]))
+            draw.rectangle(bbox, outline='red')
+            draw.text((bbox[0][0]+60, bbox[0][1]), text, font=font, fill='red')
+
+
+    preds.append(text.split()[0])
+
+    # Save the output image with bounding boxes and OCR result
+    out_path = os.path.join(output_path, f'{filename}_processed.jpg')
+    image.save(out_path)
+
+# calculate accuracy
+correct_preds = sum([1 if str(label) == str(pred) else 0 for label, pred in zip(labels, preds)])
+print(labels)
+print(preds)
+accuracy = correct_preds / len(labels) * 100
+print(f"Accuracy: {accuracy}%")
+
+char_accuracy = 0
+for l, p in zip(labels, preds):
+    corrects = 0
+    for i in range(len(l)):
+        try:
+            if l[i] == p[i]:
+                corrects += 1
+        except:
+            continue
+    per_accuracy = corrects / len(l)
+    char_accuracy += per_accuracy
+char_accuracy = char_accuracy / len(labels)
+print("Character Based Accuracy : ", char_accuracy)